By Pascal Fintoni, AI & Digital Marketing Strategist
The businesses getting real value from AI agents are the ones making them earn it first.
In 1994, on my very first website project in London, I did something faintly ridiculous. I had just been given an email address, and I did not know a single other person who had one. So, I emailed myself, just to see if the thing actually worked. It felt absurd at the time.
Looking back, it was the right instinct. I did not trust a brand new tool with anything real until I had tested it on something that could not go wrong. Thirty-two years on, I find myself giving business owners exactly the same advice, except the tool now is an AI agent, and what they want to hand it matters a great deal more than a test email.
Give it a proper brief first
Agents are the live conversation in every boardroom and café I sit in at the moment. Business owners are worn out by repetitive admin and want an assistant that can handle client follow ups or data processing on its own.
I understand the appeal completely but let’s be careful. People are reaching for full autonomy before they have learned to brief a basic model properly.
So let’s go back a few steps.
Think of a platform like ChatGPT or Copilot as a village full of specialists. If you hired someone new, you would not hand them the keys to the front door on their first morning and walk off. You would give them a clear brief, watch how they worked, and check what they produced before trusting them with anything unsupervised. A digital assistant needs the same discipline, if anything more of it, because it cannot fill in the gaps the way an experienced colleague would.
The ABCD brief
The method I teach in every masterclass is simple enough to use today. Tell the assistant exactly who it is acting as, with real expertise, not a vague job title. Break the task into numbered steps rather than one long paragraph. Specify the content format you want back. Then hand over the detail a new colleague would need: your customers, your tone, your limits. Try it this week:
“Acting as an experienced operations consultant, review how I currently handle [a specific repetitive task]. First map out every step in order. Then flag which steps are predictable and rule-based. Finally, tell me which of those would genuinely need an AI agent rather than simple software, with your recommendations broken into numbered steps. Here is the context: [your team size, the tools you already use, and anything that must never go wrong].”
Run that brief in a closed, private space first, on data that does not matter, before you let it anywhere near data that does.
Gartner’s own research backs this up. The firm expects more than 40 per cent of agentic AI projects to be cancelled by the end of 2027, and the reasons it gives are not about what the technology can do. They point to cost overruns, unclear payoff, and weak risk controls. Most of them, in other words, were never properly briefed before they were set loose.
An agent is not a shortcut around good management. It is a new kind of team member, and it deserves the same probation any new hire would get. Test it somewhere small first. Only then give it the keys.
Get the discipline right, and agents genuinely do free up hours every week. Skip it, and you have not automated your business. You have just found a much faster way to make the same mistakes.
To your success!
Pascal Fintoni is an AI and digital marketing strategist with 30+ years of experience working with business leaders on responsible AI adoption, content strategy, and digital marketing.
www.pascalfintoni.com